Behavioral Economics. Lecture 10

Transcription

1 Behavioral Economics. Lecture 10 Xavier Gabaix April 15, 2004

2 1 Hyperbolic discounting Luttmer and Mariotti (JPE 2003) hyperbolics does not make much difference/improvement over exponential discounting. Gruber and Koszegi rational cigarettes behavior: exponential and hyperbolics have similar consumption behavior The main difference between exponentials and hyperbolics is the predilection of hyperbolics to hoard illiquid assets. This is corroborated by evidence.

3 2 Gul Pesendorfer Self Control and the Theory of Consumption W ({c t, m t }) = δ t (u (c t ) + v (c t ) v (m t )) t 0 where c t is the actual consumption and m t is the maximum possible consumption. Assumptions: u + v concave, v convex

4 Big gain: no dynamic inconsistency People don t like dynamic inconsistency because of: technical difficulties involved their philosophical stance problems with doing welfare analysis

5 2.1 Preference reversals Start with (c, c, c,...) At t = 1 you can choose between α at τ or β at τ + 1 where β > α. Does the agent prefer β?

6 If τ = 1 then agent chooses β iff δ (u (c) + v (c) v (c + α)) + δ 2 (u (c + β) + v (c + β) v (c + β)) δ (u (c + α) + v (c + α) v (c + α)) + δ 2 (u (c) + v (c) v (c + β)) If I could not commit to the plan at τ = 2, 3,... than the condition is the same except for the multiplicative factor δ τ 1. If I can commit then there will be no temptation and the condition is δ τ u (c) + δ τ+1 u (c + β) δ τ u (c + α) + δ τ+1 u (c) Now, if I can commit to the plan at t = 1 then there might be a preference reversal (we have three free parameters v (c + α), v (c + β), v (c) to fit two inequalities).

7 2.2 Time preferences and steady state Euler equation If I increase consumption from c t to c t + dε and offset with decrease from c t+1 to c t+1 (1 + r) dε then m t+1 also decreases by (1 + r) dε and I gain ( V = u (ct )+v (c t )+δ (1 + r) u (c t+1 ) (1 + r) v (c t+1 ) + (1 + ε

8 Thus V ε = 0 gives u (c t ) + v (c t ) r = u (ct+1 ) + v (c t+1 ) v (m t+1 ) δ

9 Take an economy with different types temptation. (u, λ; v, δ) i=1,...,n where λv is now n Total endowment w = i=1 c it. Take u (c) = ln c and v (c) = c We get 1 + r t+1 = 1 c it+1 c 1 + λ i it 1 + λ i λ i δ

10 In steady state c it = c i and r t = r,and 1 c i + λ i r = 1ci + λ i λ i δ hence c i = δ (1 + r) 1 λ i Call γ i =λ 1. Then c i i = [δ (1 + r) 1] γ i = αγ i for appropriate α Then w = c i = α ( γ i ) Hence γ i c i = γi w

11 Gul Pesendorfer is very unexplored model, and many people like it more than hyperbolics. Does it lead to different results than hyperbolics? It s not well understood. Frederick, Loewenstein, and O Donoghue (JEL 2002) review of time discounting.

12 3 Macro 3.1 Inflation Nominal illusion Fact. Most people don t master the difference between nominal and real quantities

13 Modigliani Cohn hypothesis. Impact of nominal illusions on stock market prices Take a rational model when dividend is discounted at rate r+π (where r is interest rate and π is risk premium). Gordon formula p 1 = D r + π g where g is rate of growth of dividends. Take g = 0. D p to If people have nominal illusions then they compare dividend yield the nominal interest rate r + i (where i i s inflation). [note that bond yield usually includes inflation]

14 If the representative agent is victim of this illusion, then the required premium on stocks will be r + π = r + i + β where β is some rule of thumb risk premium So an econometrician measures π = i+β and obtain risk premium/excess return that is increasing with inflation. If all agents are rational the measured π is independent of inflation. If some agents are boundedly rational then you expect for some γ (0, 1). π = γi + α

15 Thus stock market is down when inflation is high. Other explanations: high inflation may mean other things going badly in the economy. Does the Modigliani Cohn hypothesis hold? Evidence is inconclusive The latest attempt (Campbell and Vuolteenaho 2003) suggest that the MC hypothesis does hold.

16 Irving Fisher effects? If the Fisher hypothesis holds then nominal interest rates R t = r + i t for some constant real productivity r and the real interest rate is independent of inflation. In a very behavioral world with nominal illusion we can have 0 coefficient on inflation, or and the real interest rate equals R t = α + γi t r t = α (1 γ) i t Thus r t is low when inflation is high. Empirically, mixed evidence.

17 3.1.2 Other behavioral dimensions of inflation Aversion to nominal wage cuts (Akerlof, Dickens, and Perry, Brookings 1996). They show a histogram of nominal wage changes: big mass at 0%, 1%, 2%, etc. You also have some firms at 4% or 5% but you very little mass immediately below 0. Thus, firms really don t like small nominal wage cuts. This is an argument against 0 inflation. Unemployment rate is will be higher at 0% inflation, as we hit the constraint of (almost) no nominal wage cuts. There is also some evidence: Switzerland used to have 0% inflation and many things were going badly.

18 Akerlof, Dickens, and Perry, Brookings 1996 model that, and provide evidence.

19 Real costs of inflation, for lowish inflation (between 0 and 10%) Many of the traditional costs are likely to be small: Allais Baumol Tobin shoe leather cost of going to bank: They are likely to be small. cf Calibration by Lucas (Econometrica, 2000). Menu cost of changing prices and producing new menus. Price distorsions induced by inflation volatility (e.g. Bénabou) Some costs due to bounded rationality are likely to be bigger: Thinking costs: It s a hassle to have to handle inflation all the time.

20 If people are victims of money illusion, then very important prices are distored (e.g. stocks: Modigliani Cohn, and bonds: if the Fisher hypothesis doesn t hold) For very low inflation (<1%): The aversion to nominal wage cut becomes a very big issue, and probably the major cost of inflation.